NBER WORKING PAPER SERIES BIRTHPLACE DIVERSITY AND ECONOMIC PROSPERITY. Alberto Alesina Johann Harnoss Hillel Rapoport

Transcription

1 NBER WORKING PAPER SERIES BIRTHPLACE DIVERSITY AND ECONOMIC PROSPERITY Alberto Alesina Johann Harnoss Hillel Rapoport Working Paper NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts Avenue Cambridge, MA January 2013 We thank Frédéric Docquier for sharing his immigration data with us. We also thank Amandine Aubry, Simone Bertoli, François Bourguignon, Jesús Fernández-Huertas Moraga, Frédéric Jouneau, Joao Santos-Silva, Jacques Silber, Silvana Tenreyro, and participants at the 5th AFD-World Bank Conference on Migration and Development, Paris, June 2012, and the 1st CEMIR conference at CESifo, Munich, December 2012, for comments on a previous draft. The views expressed herein are those of the authors and do not necessarily reflect the views of the National Bureau of Economic Research. NBER working papers are circulated for discussion and comment purposes. They have not been peerreviewed or been subject to the review by the NBER Board of Directors that accompanies official NBER publications by Alberto Alesina, Johann Harnoss, and Hillel Rapoport. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including notice, is given to the source.

2 Birthplace Diversity and Economic Prosperity Alberto Alesina, Johann Harnoss, and Hillel Rapoport NBER Working Paper No January 2013 JEL No. F22,F43,O1,O4 ABSTRACT The diversity of people has economic costs and benefits. Using recent immigration data from 195 countries, we propose an index of diversity based on people s birthplaces. This new index is decomposed in a size (share of foreign born) and a variety (diversity of immigrants) component and is available for 1990 and 2000 and for the overall as well as for the high (workers with college education) and low-skill fractions of the workforce. We show that birthplace diversity is largely uncorrelated with ethnic and linguistic fractionalization and that unlike fractionalization it is positively related to economic development even after controlling for education, institutions, ethnic and linguistic fractionalization, trade openness, geography, market size and origin-effects. This positive association appears particularly strong for the diversity of skilled immigrants in richer countries. We make progress towards addressing endogeneity by specifying a gravity model to predict the diversity of immigration based on exogenous bilateral variables. The results are robust across various OLS and 2SLS specifications. Alberto Alesina Department of Economics Harvard University Littauer Center 210 Cambridge, MA and IGIER and also NBER aalesina@harvard.edu Hillel Rapoport Department of Economics Bar Ilan University Ramat Gan, Israel and EQUIPPE, University of Lille and also Center for International Development, Harvard University hillel.rapoport@mail.biu.ac.il Johann Harnoss EQUIPPE, University of Lille and Harvard University johann.harnoss@post.harvard.edu

3 1 Introduction Foreign-born individuals now represent on average ten percent of the workforce of the OECD, a threefold increase since 1960 and a twofold increase over the past twenty years. 1 This growing diversity in terms of country of birth may have far reaching economic implications. Economic theory suggests that higher diversity could lead to beneficial skill complementarities in certain production processes, but also to inefficiencies via higher transaction costs due to mistrust and lack of social cohesion. 2 The empirical literature has so far focused on ethnic and linguistic fractionalization, which were shown to exert negative effects on economic growth in cross-country comparisons (Easterly and Levine, 1997, Collier 2001, Alesina et al., 2003, 2012), and on genetic diversity. 3 In this paper we look at the relationship between diversity and development using a new perspective that focuses on the diversity arising from people s birthplaces. Albeit loosely linked through immigration, ethnic and birthplace diversity are empirically (perhaps surprisingly) almost completely uncorrelated. Conceptually, ethnic and birthplace diversity also differ as people born in different countries are likely to have been educated in different school systems, learned different skills, and speak different languages; once gathered in a single country, first-generation immigrants arguably form a more diverse group than second and third-generation immigrants (or than people of different ethnicities) that grew up speaking the same language and, more often than not, learned from each other inside or outside of school: the melting pot does indeed melt! This paper makes three contributions. First, we construct and discuss the properties of a new index of birthplace diversity. We build indicators of diversity for the workforce of 195 countries in 1990 and 2000, disaggregated by skill/education level, and computed both for the workforce as a whole and for its foreign-born component. In so doing, we add a new dimension to the diversity literature, which already includes measures of ethnic, linguistic, cultural, and genetic diversity. Second, we investigate the relationship between birthplace diversity and economic development empirically. Using OLS for a large sample of countries, we find that in contrast to ethno-linguistic fractionalization, birthplace diversity is positively related to productivity in a rich model where we control for a large range of potential confounding factors such as education, institutions, trade openness and trade diversity, ethnic and linguistic fraction- 1 See Ozden et al. (2011) for a picture of the global evolution of international migration over the last fifty years. The figures for high-skill migration are even more impressive, with a twofold increase during the 1990s only (see Docquier and Rapoport, 2012). 2 As discussed in Alesina and La Ferrara (2005). 3 Spolaore and Wacziarg (2009) find genetic distance to be a strong predictor of income differences between pairs of countries and conclude that genetic distance works as a barrier to knowledge diffusion and technology adoption. Most recently, Ashraf and Galor (2012) find an inverted u-shaped relationship between genetic diversity within a population and productivity, indicating the trade-off between beneficial forces of diversity expanding the technology frontier and detrimental ones leading to higher inefficiency due to communication and coordination problems. In another paper, Ashraf and Galor (2011) find that cultural diversity (based on World Values Survey data) is positively correlated with contemporary development and suggest that cultural diversity facilitated the transition from agricultural to industrial societies. 2

4 alization, geography, and what we term origin-effects taking into account the level of productivity in the migrants home countries. This positive effect is stronger for skilled migrants (workers with college education) in richer, more productive countries, suggesting the presence of production function effects of diversity in innovative tasks in countries closer to the technology frontier. In terms of magnitudes, increasing the diversity of immigrants by one standard deviation is correlated with a rise in long-run real income by a factor of 1.2 to 1.5. Third, we make progress towards addressing endogeneity issues arising from the fact that rich countries may attract a diverse group of immigrants because they are rich rather than becoming rich thanks to a diverse workforce. We specify a gravity model of migration to predict the diversity of a country s immigration using exogeneous geographic and cultural bilateral variables and confirm our initial findings in a range of 2SLS models. The empirical evidence on birthplace diversity and income levels is scant and, to the best of our knowledge, limited to the context of the United States. Ottaviano and Peri (2006) construct a measure of cultural diversity from using migration data on US metropolitan areas and find positive effects on the productivity of native workers as measured by their wages. More recently, Peri (2012) found positive effects of the diversity coming from immigration on the productivity of US states, a result he attributes to unskilled migrants promoting efficient task-specialization and adoption of unskilled-efficient technologies, and more so when immigration is diverse. 4 Ager and Brückner (2011) study the link between immigration, diversity and economic growth in the context of the United States about a century ago, at a time now commonly referred to as the age of mass migration (Hatton and Williamson, 1998). 5 They find that fractionalization increases output while polarization decreases it in US counties during the period Finally, a paper by Ortega and Peri (2012) developed independently from this paper also analyzes the connection between income per capita and openness to (and diversity of) trade and immigration, respectively, in a cross-section of countries. They show that in the horse race between immigration and trade to explain cross-country differences in economic performance, immigration is a clear winner. The rest of this paper proceeds as follows. Section 2 briefly discusses the channels through which birthplace diversity can affect productivity and the way it relates to other dimensions of diversity (e.g., ethnic, linguistic). Section 3 explains the construction and analytical decomposition of our birthplace diversity index; we also run initial regressions comparing ethnic fractionalization and birthplace diversity s association with economic development. In section 4 4 Ottaviano and Peri (2006) and Peri (2012) employ instrumental variables techniques to account for endogeneity. They construct a measure of predicted immigration based on geographic proximity to immigration gateways into the U.S., such as New York or Los Angeles, and rely on Card s (2001) shift share methodology to predict changes in immigration by extrapolating the local levels of immigrant communities using national level immigration rates, thus making immigration exogenous to state-level economic shocks. 5 See also Bandiera, Rasul and Viarengo (2012) and Abramitzky, Boustan and Eriksson (2012) respectively on the measurement of entry and return flows and on migrants selfselection during that period. 3

5 we investigate the relationship between birthplace diversity and income more deeply, questioning its strength and robustness to a range of alternative specifications, and confront endogeneity issues using a gravity framework in Section 5. Section 6 concludes. 2 Skill complementarities and diversity 2.1 The costs and benefits of diversity The reason why birthplace diversity could be beneficial for productivity is due to skill complementarity. People born in different places are likely to have different productive skills because they have been exposed to different experiences, different school systems, different cultures and thus have developed different perspectives that allow them to interpret and solve problems differently. These differences can be complementary and lead to higher productivity. 6 Lazear (1999a,b) proposes a model of teams of workers where diversity brings benefits via production complementarities from relevant disjoint information sets and also costs via barriers to communication; with decreasing marginal benefits and increasing marginal costs, this suggests that there is an optimal degree of diversity. 7 Hong and Page (2001) see two sources for the heterogeneity of people s minds: cognitive differences between people s internal perspectives (their interpretation of a complex problem) as well as their heuristics (their algorithms to solve these problems). They show theoretically that, under certain conditions, a group of cognitively diverse but skill-limited workers can outperform a homogenous group of highly skilled workers. Empirical studies of diverse teams in the management and organization literature also find diversity to be a double-edged sword, with diversity (in terms of gender, education, tenure, nationality) being often beneficial for performance but also decreasing team cohesion and increasing coordination costs (see Milliken and Martins, 1996, and O Reilly et al., 1989). Specifically, in a study of the oligopolistic airline industry with observable actions and reactions, Hambrick et al. (1996) find that heterogeneous management teams react more slowly to a competitor s actions, but also yield higher market shares and profits than their homogeneous competitors. In a recent experimental study, Hoogendoorn and van Praag (2012) set up a randomized experiment in which business school students were assigned to manage a fictitious business and increase outcome metrics like market share, sales and profits of their business. The authors find that more ethnically diverse teams (defined by parents countries of birth) outperform more homogeneous ones, but only if the majority of team members 6 Alesina et al. (2000) formalize this idea using a Dixit-Stiglitz type production function where outputs increase in the variety of inputs and inputs can be interpreted as different workers. Their model thus allows for diversity to increase output without any counterbalancing costs. 7 A related argument, also brought forward by Lazear (1999b), is that diverse groups of immigrants tend to assimilate more quickly (in terms of learning the language of the majority) since they have stronger incentives to do so. 4

6 is foreign. However, the exact causal mechanism between ethnic diversity and higher performance remains unobserved in this study. Finally, a few recent studies have used firm-level data to explore the links between workers diversity and firms productivity. Brunow et al. (2012) analyze the impact of birthplace diversity on firm level productivity in Germany. Addressing endogeneity through a system GMM approach, they find that the share of foreigners has no effect on firm productivity while the diversity of foreign workers has a strong positive effect on firm performance (as does workers diversity at the regional level). These effects appear to be stronger for manufacturing and high tech industries, suggesting the presence of skill complementarities at the firm level as well as regional spill-overs from workforce diversity. Parrotta et al. (2012) use a firm level dataset of matched employee-employer records to analyze the effects of diversity in terms of skills, age and ethnicity on firm productivity. They find that while diversity in skills increases productivity, diversity in ethnicity and age has negative effects on productivity. They interpret this as showing that the costs of ethnic diversity outweigh its benefits. However and quite interestingly, they also find suggestive evidence that ethnic diversity is more valuable in problem-solving oriented tasks and in innovative industries as the effect of ethnic diversity turns less negative for white-collar workers and also turns less negative in more innovative research and development-intensive industries. Last, Boeheim, Horvath and Mayr (2012) find micro level evidence for the presence of production function complementarities using a linked dataset of Austrian firms and their workers during the period They partly address endogeneity through an IV approach and find that workers wages increase in the birthplace diversity of co-workers but decrease in the own group share. The wage effects appear stronger in subsamples for white-collar workers and workers with young tenure, again suggesting complementarities in the production of non-routine tasks and learning externalities at the firm level. At a macro level, the costs of diversity have been established theoretically and empirically, in particular for ethno-linguistic differences. These studies mostly began with Easterly and Levine (1997), who show that ethnic fragmentation is associated with lower economic growth, specifically in Africa. Collier (1999, 2001) adds that ethnic fractionalization is less detrimental in the presence of democratic institutions, which enable different groups to mediate conflicts on the provision of public goods and create social cohesion, although it is unclear whether it is just the level of per capita income or democracy that matters since the two are highly correlated. Alesina and La Ferrara (2000, 2002) stress the role of trust, showing that individuals in racially diverse cities in the US participate less frequently in social activities and trust their neighbors to a lesser degree, while overall trust in political institutions remains unchanged. The authors also find evidence that preferences for redistribution are lower in racially diverse communities. This also extends to the provision of productive public goods (Alesina, Baqir and Easterly, 1999). Alesina, Michalopoulos and Papaioannou (2012) stress the inequality dimension of ethnic diversity (i.e., it is the interplay between ethnic fractionalization and ethnic inequality that leads to conflict) while Esteban, Mayoral and Ray (2012) distinguish conflicts over 5

7 public and private goods and find polarization to correlate positively with conflict on the former, and fractionalization to correlate positively with the latter (see also Esteban and Ray, 2011). 2.2 Which diversity? A population s diversity is commonly measured by fractionalization (Easterly and Levine 1997, Alesina et al., 2003, Fearon, 2003) and polarization indices (Esteban and Ray, 1994, Reynal-Querol 2002). Ethnic fractionalization measures are unable to distinguish, for example, between a first- and second generation Italians in the US, neither can they distinguish between these two and two German first- and second-generation immigrants. All these people would be labeled ethnically as White/Caucasian or European Descent. A linguistic fractionalization measure would be more accurate in separating language groups but would equally fail to distinguish between first and second-generation immigrants. Still, these measures capture very fundamental (and often intergenerationally transmitted) differences between people. A dimension of diversity among people that remains largely understudied is the diversity caused by differences in people s country of birth. If early preworking age years are formative for one s own values, perspectives and skills (note that the emphasis is on skills, not on level or quality of education), these differences last a lifetime and may serve as variation to be explored for economic analysis. Shaped by different education systems and social values, this type of diversity is more likely to result in production function complementarities than differences in skin color or language spoken at home. To explore this dimension of diversity, this paper introduces a new diversity index which is more likely to be closer to the correct one when we try to explain skill complementarity: diversity in people s birthplaces. 3 An index of birthplace diversity 3.1 Data Our computation of birthplace diversity indices relies on the World Bank sponsored Docquier, Ozden, Parsons and Artuc (2012) (henceforth DOPA) data set, the last update of the Docquier and Marfouk (2006) data set which has been extended to include bilateral data on immigration by country of birth, skill category (skilled v. unskilled, the former having college education) and gender, for 195 sending/receiving countries in 1990 and The main addition to the previous versions is that the data set now captures South-South migration based mainly on observations and, occasionally, on estimated data points (for the skill structure). Immigrants are defined as foreign-born individuals aged 25 or more at census or survey date. The DOPA data set, therefore, allows for characterizing the size, origin-mix, and skill structure of the foreign-born labor 6

8 force. 8 Before we turn to birthplace diversity indices, we briefly discuss a few caveats regarding illegal immigration, the definition of an immigrant, and the skill structure. First, the fact that illegal immigration is not accounted for in most censuses is a clear limitation. In our case, this limitation is mitigated by the fact that some countries such as the United States try to account for illegal immigration in their census and, more importantly, by the fact that we use data on immigration stocks, not flows. Indeed, most illegal migrants eventually become legalized or return, meaning that even if illegal migrants represent a large fraction of total flows, they generally remain a small fraction of the immigrant stock. A second caveat concerns the very definition of an immigrant as a foreign-born individual. According to this definition, a 2-year old child immigrating with her parents will be classified as an immigrant; however, that person will grow up, socialize and go to school in the home country, thus limiting the extent of foreign skills that he or she can contribute. 9 Finally and on a related note, an individual is considered skilled independently of whether college education was obtained in the home, host, or a third country, meaning that the category skilled workers may be very heterogeneous in terms of human capital quality; we partly address this issue by controlling for what we call origin-effects (see below). 3.2 Measuring birthplace diversity: a decomposition We base our birthplace diversity measure on the Herfindahl diversity index. Let s i refer to the share in the total population of individuals born in country i with i = 1,..., I. In particular, i = 1 refers to those individuals who were not born abroad and are thus natives. The fractionalization index Div pop may be expressed as Div pop = I s i (1 s i ) (1) i=1 This index measures the probability that two individuals drawn randomly from the entire population have two different countries of birth. Since I i=1 s j = 1, equation (1) may also be written as the commonly known Herfindahl index Div pop = 1 I (s i ) 2 (2) 8 A small number of (usually small) countries is reported as having zero immigrant stocks and, on few occasions, the authors do not provide a split by skill level. Due to these missing information, we are unable to compute diversity indices for some countries: for 1990, 13 (overall), 33 (skilled) and 14 (unskilled) countries are missing and, for 2000, 8 (overall), 27 (skilled) and 11 (unskilled) countries are missing (see Table A3 in the Appendix for the details). 9 Beine, Docquier and Rapoport (2007) use age at migration as a proxy for where education was acquired. They propose migration figures corrected for age of entry for 1990 and 2000 for high-skill immigrants only. i=1 7

9 A certain level of say moderate Div pop may come from a relatively small but very diverse pool of immigrants, or by a relatively homogenous but large fraction of immigrants in the population. It is useful then to develop indices which can highlight these differences. We therefore decompose our diversity index into a Div between and a Div within component. We define Div between as the diversity from immigration, irrespective of further country of origin-differences. Div within then captures all residual diversity from differences between immigrants only. If we assume that all immigrants are born in one country i = 2 so that s 1 + s 2 = 1, then using (1) we can define: Div between = s 1 (1 s 1 ) + (1 s 1 ) s 1 (3) This essentially calculates the Div pop index assuming that all migrants can be grouped into one category (1 s 1 ) - thus excluding all diversity contributed by the fact that migrants tend to come from more than one origin country. We rewrite (4) to include Div between as follows: Div pop = s 1 (1 s 1 ) + (1 s 1 ) s 1 + I [s i (1 s i )] (1 s 1 ) s 1 (4) i=2 Since I i=2 s i = (1 s 1 ), (4) simplifies to We can now define Div pop = 2 s 1 (1 s 1 ) + Div within = I [s i ((1 s i ) s 1 )] (5) i=2 I [s i ((1 s i ) s 1 )] (6) i=2 so that Div pop is composed of two parts, Div between and Div within : Div pop = Div between + Div within (7) However, this separation is not yet fully satisfying, since it does not separate clearly between size and variety effects: Div within still depends on s 1 - the share of natives since I i=2 s i = (1 s 1 ). We thus rewrite the Div within component so that it does not depend on s 1. We achieve this by defining s j as the share of immigrants from country j in the total population of immigrants. It follows that s j = si (1 s where s 1) 1 is the share of natives (i = 1). We thus re-scale Div within using (6): [ ] I s i Div within = (1 s 1 ) ((1 s i) s 1 ) (1 s 1 ) 2 (8) (1 s 1 i=2 8

10 and simplify to: Since s j = Div within = [ ] I s [ i (1 s 1 ) 1 s ] i (1 s 1 ) 2 (9) (1 s 1 i=2 si (1 s, then: 1) Div within = J j=1 [ ] s j (1 s j ) (1 s 1 ) 2 (10) Our result has a very intuitive interpretation: since J j=1 [ ] s j (1 s j ) is basically (1) but applied to the population of immigrants, it is essentially a diversity index of immigrants only, irrespective of the natives. We thus define: And rewrite (7) Div Mig = J j=1 [ ] s j (1 s j ) (11) Div pop = Div between + (1 s 1 ) 2 Div Mig (12) where (1 s 1 ) 2 has an intuitive interpretation as scale parameter for Div Mig. We can then rewrite (12) in terms of s F, the share of immigrants (defined as foreign-born) and define s F = (1 s 1 ): Div pop = 2 s F (1 s F ) + (s F ) 2 Div Mig (13) We have thus an expression of Div pop purely as a function of the size and diversity of immigration. 3.3 Descriptive statistics We now compare the properties of these size and variety measures with each other and with an index of ethnic fractionalization (taken from Alesina et al., 2003). A first visual overview (see Figure 1) shows that ethnic fractionalization and Div pop differ considerably: where ethnic differences are high, diversity of origins is quite low (e.g., in all African countries with few exceptions and in Central and South-East Asia). In turn, Div pop is high in North-America, Europe, Australia and some Arab countries, while ethnic differences are much more modest. This is also reflected in the low bilateral correlation between them, as can seen from Table 1 and Figure 2. Birthplace diversity (Div pop ) is highly correlated (+0.98) with the share of immigration (s F ). This explains the very low overall diversity in large countries such as China, India and Brazil with few foreigners relative to overall population. To complement the picture, Figure 9

11 1 also displays a third map showing the diversity of migrants, Div Mig. Interestingly, it is quite unrelated to the overall diversity of a country s population. North-America, Europe and even some Eastern European countries exhibit a very high diversity of immigrants. (see Table 2 for regional comparisons). Latin American countries, some African countries, China and Russia have intermediate diversity, and Pakistan, India, Indonesia, Iran and many (but not all) African countries are not very diverse in terms of their immigrants. Table 3 presents fractionalization and diversity scores for a selection of countries. The pure variety of birthplaces is highest in many rich countries: Canada, Italy, Israel, Germany, Australia and the UK all have diversity of immigrants at about.9. The United States rank only 20 in a list of the most diverse immigrant countries (at.92) due to its relatively low diversity of unskilled workers (0.84). In terms of skilled diversity, the USA is the second most diverse country together with Italy (with both countries at.97). However, the foreign-born represent only 1.8 percent of the skilled labor force in Italy versus 11 percent in the US. The diversity of skilled immigrants is also high in Germany, UK, France, Spain, and Canada as well as in countries such as Argentina, Brazil and Qatar. Countries with lowest overall diversity are Pakistan, Bangladesh, Nepal, Iran (all lower than.1). We can observe some clear neighborhood effects: Ireland s diversity (.55 overall,.44 for the unskilled and.67 for the skilled) is still quite low due to dominant immigration from the UK. Similar effects can be observed for, e.g., Switzerland, Austria, and also Australia where UK immigration is very large compared to other immigrant groups. Nepal and Sri Lanka experience immigration from a very large neighboring country, India. Neighboring effects are strongest for immigration of unskilled workers, while diversity of skilled workers is in general higher than the diversity of unskilled workers. This is consistent with migrants self-selection being driven by net-of-migration-costs wage differentials, where low migration costs (due to short distances and high networks) mostly affect low-skill migration. 10 The correlation between ethnic fractionalization and birthplace diversity of immigrants is very low (-.01 overall) and even negative at -.2 (in 2000) for skilled immigrants (see Table 1). This confirms that immigrants diversity and ethnic fractionalization are indeed quite different measures. Besides, the correlation between s F and Div Mig is surprisingly low (see Table 1), especially for skilled immigrations: the diversity of immigrants is basically uncorrelated with the size of immigration (+.06) for skilled and overall immigration, and is only slightly positively correlated for unskilled immigrants (+.09 in 2000). This suggests that the size of immigration (size) and its diversity are largely independent. This observation holds irrespective of country size. However, there is a positive correlation between immigration s size and diversity when we look at first differences: this suggests that, at least for the 1990s, the variety of immigrants rises when a country experiences an inflow of immigrants. This holds true in particular for skilled workers (bilateral correla- 10 See Munshi (2003), McKenzie and Rapoport (2010) and Bertoli (2010) for micro evidence, and Grogger and Hanson (2011) and Beine, Docquier and Ozden (2011) for macro evidence. 10

12 tion +.22) but, interestingly, does not hold for unskilled immigrants (bilateral correlation +.06). This is most likely due to the role of diaspora/immigrant networks, which tends to reduce migration costs mainly for unskilled workers. The experience of the United States symbolizes this effect: the great Mexican migration explains that the share of unskilled immigrants has increased and at the same time diversity has decreased as it became more polarized. This pattern, however, hides considerable heterogeneity: for example, Malaysia has reduced its share of skilled immigrants (primarily a technical effect due to higher domestic educational attainment and thus a broader base) and also increased its diversity of skilled immigrants, whereas Uganda or the Czech Republic saw higher skilled immigration, but lower diversity. Skilled and unskilled diversity are highly correlated overall, with correlation coefficient of for the year 2000 (see the last panel of Figure 2). 11 However, there are some interesting deviations from this relationship: as already stated, the United States have a higher diversity of skilled immigrants than of unskilled immigrants, primarily due to the large inflow of Mexican immigrants that dominates the group of unskilled immigrants. The same holds true for many other countries, such as Ireland, Malaysia, and, to a lower extent, Singapore. All these countries have a large neighboring country that tends to draw a high share of unskilled immigrants, thus lowering the diversity of unskilled immigrants relatively to skilled ones. 3.4 Ethnic fractionalization and birthplace diversity The data analysis of the previous section highlights that, broadly speaking, birthplace diversity and the diversity among immigrants is high in rich countries while ethnic diversity is larger in poor countries. In this section we formalize these correlations. We run the following simple model: ln y kt = α + β 1 birthplace diversity + β 2 fractionalization + e (14) We expect the two coefficients β 1 and β 2 to have different signs and to be statistically different from each other. For birthplace diversity, we use Div pop in some specifications and replace it later by our size and variety components s F and Div mig. For fractionalization, we use Alesina et al. s (2003) ethnic and linguistic fractionalization measures. Table A1 in the Appendix details the definitions and sources for all our variables. Our aim is to show a simple bilateral correlation, irrespective of other confounding effects. We use the broadest sample of 135 countries with data for GDP, TFP, both fractionalization measures (ethnic and linguistic) and all diversity variables on a cross section for the year 2000 using heteroskedasticity-robust standard errors. 11 Skilled and unskilled migrations are typically highly correlated, with skilled immigration Granger causing unskilled immigration (Gibson and McKenzie, 2011) through chain migration and network effects. On the latter and their differential effects at different skill levels, see McKenzie and Rapoport (2010). 11

13 We present results separately for GDP/capita and TFP/capita in Table 4. The results confirm our main hypothesis, namely, that measures of ethnic and linguistic fractionalization enter negatively whereas birthplace diversity (Div pop ) enters positively in all models. We also find that fractionalization turns more negative (increases in magnitude) once birthplace diversity is controlled for this change in magnitude is not significant, however. Nevertheless, this implies that fractionalization may tend to capture (some) positive effects of birthplace diversity if used in isolation. Most importantly, we establish that our size and variety components both relate positively to economic development and remain highly significant (mostly at the 1% level) when entered jointly. 4 Empirical analysis 4.1 Model and data In this section we empirically investigate the relationship between birthplace diversity and economic development. We specify the following model where our dependent variable y is a country s income or productivity per capita: 12 ln y kt = α + β 1 diversity migrants skt + β 2 share immigration skt + β 3 origin effects skt + β 4 years of schooling + β 5 market size controls + β 6 Γ kt + β 7 k + β 8 Φ kt + β 9 Ψ kt + η t + e (15) As we detail below, Γ kt is a vector of geographic characteristics, k is a vector of fractionalization measures, Φ kt is a control for institutional development, Ψ kt is a vector of controls for trade openness and trade diversity, and η t is a time fixed-effect. We use indices s for skill groups (s=overall, skilled, unskilled), t for time (1990, 2000) and k for countries. The results from our decomposition as well as our previous analysis point to the need to separate the share of foreigners, s F, and the diversity of immigrants, Div Mig, to isolate size and variety effects. Our empirical specification thus includes the share and the diversity of immigrants. 13 An alternative specification which interacts size and variety shows consistently high positive estimates for 12 See Table A1 in the Appendix for details on the definitions and sources for all variables. 13 Note that as a robustness check we also ran a model that directly implements equation (13) using a log-linear specification, with qualitatively similar results for our main variable of interest, Div mig. As can be seen, s F enters four times in equation (13): as a linear term, a quadratic term, an inverse term (1 s F ) and an interaction term (with Div mig ). All these terms are obviously collinear. A log-linear specification collapses these terms to just two: a linear term and an inverse term, which are actually the same variable interpreted differently (s F being the share of foreign-born and 1 s F the share of native-born), and so the regression should be collapsed to just two variables, s F and Div mig, which is what we are doing. 12

14 the interaction, but suffers from very high multicollinearity (correlation +.84 for size and the interaction term). We thus proceed to evaluate the marginal effects of size and variety at the means of the respective variables. We also run various robustness checks (see Table 10) to ensure that our results are robust to the exclusion of small countries and of countries with very low immigration. We control for a very wide range of potential confounding factors. Our model follows the literature in that it includes standard controls (such as education via years of schooling, market potential via population and area sizes, and a landlocked dummy) to which we add a series of controls entered in groups for trade structure, fractionalization, geography, institutions, and what we term origin-effects. As income differences can be related to trade (Frankel and Romer, 1999) and to the quality of institutions (Rodrik et al., 2004, Glaeser et al., 2004), we control for the volume and structure of trade as well as for the level of democracy. For institutional quality we use the Polity 2 score from the Polity IV database. As trade controls we use real trade openness from PWT and also control for the structure of trade by constructing a measure of trade diversity (Herfindahl index of exports based on Feenstra et al., 2005). This diversity index is basically the goods market equivalent of birthplace diversity and allows us to test which of the two relates more robustly to income. 15 The need to control for the volume and structure of trade stems from the fact that migration and trade share common determinants, resulting in birthplace diversity possibly capturing some of the productivity effects of trade. For the fractionalization vector, we include both ethnic and linguistic fractionalization (from Alesina et al., 2003) since both tend to capture birthplace diversity related effects to some degree (see our discussions above). For the geography vector, we follow the literature and use absolute latitude (Hall and Jones, 1999, Gallup et al., 1998, Rodriguez and Rodrik, 2001, Sachs 2003, Rodrik et al., 2004), malaria intensity (Gallup et al., 1998, Sachs 2003, Rodrik et al., 2004) and the share of population living within 100km of an icefree coast (Gallup et al., 1998). We also check the robustness of our results to the use of alternative geography variables (see Section 4.3 and Table 10). Finally, we control for what we term origin-effects of diversity via a simple weighted average of the GDP and TFP per capita of the origin countries of immigrants. This is important because it is likely that immigrants coming from richer countries can more easily afford the costs of migration, resulting in countries attracting migrants from richer countries being also more diverse. If this is the case and to the extent that migrants from high-productivity countries have a stronger effect on productivity at destination than migrants from low- 14 There is substantial ambiguity when choosing a measure for trade liberalization. Measures generally fall in two camps: trade volume or trade policy measures. We use the standard and most basic measure of trade volume: real trade openness as total export and import volume over GDP in real PPP prices. Yanikkaya (2003) compares a range of openness indicators and finds trade openness to correlate most robustly with GDP growth. 15 Our definition of trade diversity follows the trade literature as (1 - trade concentration), see e.g. Kali et al. (2007) for the effect of trade concentration on income or Frankel et al. (1995) on the link between trade concentration and transportation cost reductions. 13

15 productivity countries, then the correlation between birthplace diversity and productivity may be spurious. Controlling for such origin-effects allows for focusing on the pure diversity effect of immigration. We thus end up with a highly structured model (with our key variables and 14 covariates) and a panel of 93 countries in 1990 and We interpret the coefficient on β 1 as capturing the pure variety effects of diversity, orthogonal to a wide range of potential alternative effects or channels of influence, and β 2 to capture all size effects of immigration. The mean GDP/capita in PPP is at 4,391 USD, the diversity of immigration ranges from 0.01 to 0.96, with a mean of This is somewhat higher than the mean in the overall dataset for which we have this measure (0.65), due to many small island countries with low birthplace diversity dropping from our sample. The correlation between size and variety of immigration in our dataset is for unskilled migration and for skilled migration (see the Appendix for more descriptive statistics). 4.2 OLS results We run our model using an OLS estimator with standard errors clustered at the country level to account for serial correlation of standard errors and year fixed effects to account for year-specific shocks to all countries. We use a sample of 92 countries for which there are data for all variables, which amounts to 183 observations for the years 1990 and 2000 combined. 16 The results are presented separately for GDP and TFP per capita and also for overall, skilled and unskilled diversity in Tables 5 to 11. Table 5 presents initial results for our benchmark model and for the full sample. We introduce our controls sequentially in groups and find that only the diversity of skilled migrants relates positively and significantly at the 5% level to economic income, both for GDP and TFP/capita. For GDP, this estimate turns significant once we control for geography and institutions, whereas the relationship for TFP is positive and significant throughout the specifications. We establish that skilled diversity seems to exhibit more robust positive effects than diversity of unskilled workers. The share of immigrants, the other key determinant of overall diversity, relates positively and significantly to income for unskilled immigrants only (which drives the positive relationship for overall immigrants). If there were production function effects of diversity and given our theoretical discussions in Section 2, we should find that they are stronger in a subset of economies with more advanced production processes, which are closer to the technological frontier. We thus separate our sample into countries above and below the median GDP and TFP/capita in 1970, allowing for heterogeneity in the coefficients on birthplace diversity. 17 In Table 6, we report our results for 16 We have only one observation for Central African Republic, Liberia and Trinidad & Tobago, hence our dataset contains 183, not 186 observations. The results are robust to dropping these observations. 17 We also test the same hypothesis by splitting our sample into countries with higher and lower patent intensity and find very similar results, see Table 8. 14

16 rich countries only. We essentially find the same results as in our overall sample, but the magnitudes of our estimates on skilled diversity are much higher and significance is also higher at the 1% level (for the full model). On the contrary, for the poor countries sample in Table 7, we find no significant relationship between variety of immigrants and economic development. This heterogeneity across countries indicates that skilled diversity is particularly relevant in richer countries. For robustness, we extend our split sample approach using patent data. We define average patent intensity as average number of patents granted by national patent offices to residents as a share of population. We can replicate our rich country results for a limited group of countries that is measured in terms of relative patents close to the technological frontier (in the highest quartile). Table 8 shows our results: diversity of skilled immigrants remains stable in terms of magnitude and significant at the 5% level. 4.3 Robustness We check the robustness of our results to a range of alternative specifications and subsamples. Since the geography controls are critical in any income regression, we replace our standard controls by two alternative specifications suggested by Rodriguez and Rodrik (2001). Table 9 reports the results for share of tropics (in % of land mass area) as alternative control as well as a set of three regional dummies. We find that the correlation regarding skilled diversity of immigrants is robust: it holds at 10% in the GDP model and at 5% significance in the TFP model. The magnitudes remain stable. Interestingly, in both specifications, unskilled diversity becomes more significant. This is particularly true for the specification with regional dummies. Table 10 shows the robustness of our results (for skilled diversity only) to alternative specifications and samples. Column 1 shows our baseline result as benchmark. Columns (2) and (3) show our results remain valid when weighting observations by the size of immigration (Column 2) or when we drop the 10% of countries with lowest overall immigration (in % of population) (Column 3). We can replicate our baseline results at 5 percent significance level and very similar magnitudes. Column 4 shows our results when dropping smaller countries (excluding countries below the 10th percentile of the size distribution, i.e. below 3.1 million population). Again, our results for skilled diversity remain highly significant, suggesting that it is not small countries that are driving our overall result. Column 5 limits our sample to the 23 OECD countries in our overall sample. Running our full model on this small sample, we nevertheless find our results for skilled diversity to be significant (at 5% level) for the GDP model. In the TFP model we lose significance but replicate earlier results qualitatively. Lastly, in column 6, we show results when dropping the smallest 25% of observations for skilled diversity to verify that our results are not driven by the skewed distribution of diversity and the corresponding outliers with very low diversity. Again, we can replicate our results and find them to hold in that subsample as well. 15

17 Finally, in another robustness check (see Table 11), we control for migration networks in 1960 via a diversity index of migrants in 1960 (based on data from Ozden et al., 2011). We find that our results for diversity of skilled migrants in 1990 and 2000 lose significance but remain positive (largely due to the correlation between diversity in 1960 and 1990/2000, at about +.46). However, we can replicate our rich country subsample results at 1 percent significance levels. This indicates that lagged diversity effects do not drive our main results. 18 How to interpret this result? A descriptive analysis reveals that while in our sample both Div mig and s F have increased between 1960 and 2000, these changes do not correlate significantly with changes in GDP per capita nor do they correlate significantly one with the other. In other words, countries have become more diverse on both the size and variety dimensions of diversity but these two dimensions do not interact very much. In addition, countries that had a diverse workforce in 1960 are the ones which have a large and diverse second and third generation of immigrants today. This type of diversity somehow combines the presumably favorable effects of birthplace diversity with the more negative effects of ethnic diversity; it is therefore not so surprising that birthplace diversity in 1960 does not seem to affect the level of productivity today nor that it does not wash out the positive productivity effects of current birthplace diversity. 5 Identification In this section we discuss and go part of the way towards addressing endogeneity concerns. A first major concern is that richer countries could (and actually do) attract a larger flow of immigrants coming from a wider range of origin countries. A descriptive analysis of our data shows however that the diversity of immigration did not increase with economic growth during the period : the bivariate correlation between changes in income (in real GDP/capita) and changes in skilled diversity is extremely low at This coefficient is larger by a factor of 5 for the correlation between changes in the share of immigrants and growth over the same period, suggesting that reverse causality is a priori more a concern for the size than for the diversity of immigration. Omitted variables present another potential source of endogeneity. We addressed this second concern by controlling for a large range of factors. In particular, we accounted for the trade openness of a country and for the diversity of its trade partners since it is plausible that more outward-oriented countries would be more open to both trade and immigration. 19 Still, there are certainly remaining 18 Interestingly, when excluding small countries (with population lower than 3.1 million, i.e., countries below the 10th percentile of the population size distribution) and also the countries with very low foreign-born populations (with immigration shares lower than 1.25%, i.e., countries below the 25th percentile on the immigration share distribution), the results for the rich countries subsample remain stable while the full sample results become significant and positive as well. These results are available upon request. 19 Interestingly, this correlation between the two is very low in our dataset at for diversity of skilled immigrants and real trade openness and for diversity of immigration and trade diversity. 16

18 factors that govern the joint pattern of migration and productivity (e.g., technological progress affecting transport and communication costs). We therefore rely on an instrumentation procedure to further address these concerns. 5.1 A gravity model of migration and diversity Bilateral migration the basis for our diversity measure is determined by various economic, political, cultural and geographic factors. The trade (e.g., Tinbergen, 1962, Frankel and Romer, 1999) and migration literatures (e.g., Grogger and Hanson, 2011, Beine et al., 2011) have developed approaches to predict trade/migration aggregate flows as a function of these bilateral determinants based on the well-known gravity model. However, since we focus on income levels and productivity, we cannot use the full set of standard bilateral variables in our first-stage estimation. In particular, we cannot use the standard economic (such as absolute or relative income differences, levels of economic inequality) and political factors (such as visa policies, or differences in the quality of institutions) that usually enter in a gravity equation as this would create an endogeneity problem. Hence, we have to rely exclusively on cultural and geographic bilateral determinants that are not directly related to economic development. Our objective is to generate predicted bilateral migration stocks that will then serve as input for equation (11) and allow us to calculate an index of predicted Div mig. We thus specify the following gravity model for migration, where y ikst is the migration stock from origin country i to destination country k for immigrants of skill level s in year t: 20 y ikst = α + β 1 P OP ULAT ION kt + β 2 DIST ANCE ikt + β 3 BORDER ikt + β 4 LANGUAGE ikt + β 5 COLONY ikt + χ it + e The choice of our model determinants follows the standard in the literature (e.g., Lewer and van den Berg 2008, Spilimbergo 2009, Felbermayr et al. 2010, Mayda 2010, Grogger and Hanson 2011, Beine, Docquier and Schiff, 2012, Ortega and Peri, 2009 and 2012) but differs in objective. Again, our goal is not to accurately estimate elasticities of migration to different gravity forces but, rather, to predict migration with sufficient precision solely on the basis of a limited set of exogenous bilateral variables: bilateral (geodesic) distance, common border, common official language, and common colonial history, to which we add the destination country s population size. 21 Multilateral resistance effects (Anderson and Van Wincoop, 2003) are a concern for us if they introduce a bias to our estimates on these determinants. We thus include a set of origin-year 20 Again, see Table A1 in the Appendix for all variables definitions and sources. 21 We use the size of the destination country population to proxy for country size; origincountry population size is captured in the origin-year fixed effect. Given that the population size of the receiving country is partly determined by immigration, we also ran our model with population size in 1960 instead and found our results to be robust to this alternative specification. 17